openstarlab-event 0.1.24

openstarlab event modeliing package

OpenSTARLab Event Modeling package

Introduction

The OpenSTARLab Event package is the fundamental package for event modeling. It is designed to provide a simple and efficient way to train, inference, and simulate events. This package supports the data preprocessed by the OpenSTARLab PreProcessing package.

This package is continuously evolving to support future OpenSTARLab projects. If you have any suggestions or encounter any bugs, please feel free to open an issue.

Soccer Event Modeling

Table: Comparison of model performance on soccer event prediction

Note: Arrows indicate whether a higher (↑) or lower (↓) value is better.
Models are ranked by publication year. Bold values indicate the best performance (unrounded). For more details refer to our paper

Wyscout Dataset

Model (Year)	Action Acc. ↑	Action F1 ↑	Time-MAE ↓	X-MAE ↓	Y-MAE ↓	FLOPs	Num Params
MAJ	0.57	0.08	3.60	18.97	52.55	-	-
Seq2Event (2022)	0.67	0.16	3.41	7.11	15.72	112M	135K
NMSTPP (2023)	0.67	0.17	3.34	6.94	15.08	296M	121K
LEM_1 (2024)	0.67	0.17	3.07	8.34	21.44	50M	98K
LEM_3 (2024)	0.67	0.20	2.69	7.62	21.83	20M	39K
FMS (2024)	0.67	0.16	3.27	11.27	24.19	930M	782K

StatsBomb Dataset

Model (Year)	Action Acc. ↑	Action F1 ↑	Time-MAE ↓	X-MAE ↓	Y-MAE ↓	FLOPs	Num Params
MAJ	0.40	0.06	2.76	20.72	33.32	-	-
Seq2Event (2022)	0.65	0.23	2.43	7.22	6.86	4.03B	413K
NMSTPP (2023)	0.65	0.23	2.53	7.38	6.86	2.02B	217K
LEM_1 (2024)	0.65	0.24	2.23	7.36	8.21	66M	128K
LEM_3 (2024)	0.66	0.25	2.07	7.07	8.32	19M	38K
FMS (2024)	0.65	0.24	2.35	7.77	8.82	3.66B	1.29M

Installation

• Install pytorch (recommended version 2.4.0 linux pip python3.8 cuda12.1)

pip install torch torchvision torchaudio

• To install this package via PyPI

pip install openstarlab-event

• To install manually

git clone [e-mail]:open-starlab/Event.git
cd ./Event
pip install -e .

Current Features

Sports

• Event Model in Football/Soccer ⚽

RoadMap

• Release the package
• Provide pre-trained models

Other Information

Development torch version

version 2.4.0 linux pip python3.8 cuda12.1 

Developer

Calvin Yeung 💻

Keisuke Fujii 🧑‍💻